1. INTRODUCTION

To establish the next-generation short-reach optical interconnect (OI) for relieving heavy data traffic, the internal connection between supercomputers and data centers for cloud computing [1] has to be accelerated in order to dramatically increase the capacity of data switching and routing among numerous wired and wireless access networks. This demand has recently been met with the construction of data centers enabling data rates of up to 400 Gb/s based on the $8\times 50\mathrm{Gb}/\mathrm{s}$ array transmitter according to the IEEE P802.bs standard [2]. As proposed, the use of directly modulated vertical cavity surface emitting lasers (VCSELs) with more transverse modes at 850 nm, in combination with OM4 multi-mode fiber (MMF), has emerged as the most reliable and cost-effective solution for implementing high-speed rack-to-rack OI [3–5]. Due to its advantages of low power consumption, effective coupling to fiber [6], low threshold current, and high power conversion efficiency [7,8], the VCSEL has become a popular commercial transmitter for short-reach intra-data-center applications.

The modulated VCSEL, which has a broad mode spectrum output, suffers from severe modal dispersion during transmission in MMF, as limited by the product of transmission distance and allowable data rate [9]. Owing to this drawback, the study of few-mode (FM) VCSELs becomes important to meeting the speed requirement. By reducing the diameter of oxide-confined aperture, the allowable area of mode field and characteristic frequency of the VCSEL are confined to avoid the generation of high-order transverse modes, which strategically decreases the lasing mode number for delivering the FM VCSEL at some sacrifice of output power [10]. Unfortunately, as the oxide-confined aperture becomes smaller, the differential resistance of the VCSEL would inevitably increase to degrade the modulation depth under data modulating [11]. To solve this problem, a high-doping zinc-diffusion fabrication was considered to be added to the top distributed Bragg reflector (DBR) to improve the ohmic contact by reducing the mirror resistance through the suppression of free carrier absorption [12–14]. Alternatively, a thick benzocyclobutene (BCB) layer with low capacitance was employed to take over the traditional ${\mathrm{SiO}}_2$ isolator so as to ensure spatially confined carrier transportation of the VCSEL [15,16].

For device ameliorations, several remarkable works on VCSEL transmitters, with enhanced data rates at various formats for intra-data-center links, have been reported recently. Kuchta et al. modulated the VCSEL with none-return-to-zero on-off-keying (NRZ-OOK) data at 71 Gb/s for MMF transmission over 7 m [17]. The effect of VCSEL mode number on the allowable NRZ-OOK data capacity for back-to-back (BtB) transmission was reported [18]. To enable high spectral usage efficiency with other formats than NRZ-OOK, four-level pulse amplitude modulation (PAM-4), which needs only half the modulation bandwidth, was employed at a cost of increasing signal-to-noise ratio (SNR) criterion [19,20]. To date, the modulation of PAM-4 onto an unpackaged VCSEL chip has achieved a 100 Gb/s transmission over 100 m MMF with pre-emphasis and offline equalization at the transmitter and receiver ends, respectively [21]. To further exceed the data rate at the maintained modulating bandwidth, quadrature amplitude modulation orthogonal frequency division multiplexing (QAM-OFDM) has lately emerged as a way to modulate the VCSEL; it enables a higher spectral bandwidth usage efficiency compared to other formats [22,23]. For example, Puerta et al. demonstrated a multi-carrierless amplitude phase over 100 m MMF at 40.6 Gb/s by using an 850 nm VCSEL under the assistance of a discrete multi-tone (DMT) modulation technique [24]. Lu et al. achieved the modulation of a zinc-diffused, single-mode 850 nm VCSEL at 50 Gb/s with an analog bandwidth of only 16 GHz under the assistance of the DMT modulation technique [25]. Up to now, surveys on modulating FM zinc-diffused VCSELs for QAM-OFDM transmission in MMF over a distance of 100 m or beyond are still an uncertain issue to be approached for the short-reach intra-data-center applications.

In this work, the beyond 100 Gb/s direct modulation of a homemade FM zinc-diffused VCSEL chip at 850 nm is demonstrated for pre-leveled 16-QAM OFDM data transmission over 100 m long OM4 MMF. The optimized characteristics, including power-to-current-voltage (L-I-V) curve, normalized modulation output, and relative intensity noise (RIN) of the FM VCSEL chip are discussed. The maximal allowable transmission capacity of the FM VCSEL chip carrying 16-QAM OFDM data is implemented. The BtB and 100 m long OM4 MMF transmission performance, including error vector magnitude (EVM), SNR, bit error rate (BER), and receiving power penalty are compared and then optimized with subcarrier amplitude pre-leveling. The ultimate transmission data rates of 100 Gb/s for BtB and 92 Gb/s for 100 m MMF at a spectral-bandwidth-usage efficiency of 4 bit/s/Hz are performed as well.

2. EXPERIMENTAL SETUP

A. Fabrication Procedure of FM VCSEL

In MMF transmission, the modal-dispersion-induced group delay between the lowest- and highest-order transverse modes of the VCSEL seriously distorts the carried data steam in the time domain. Obviously, increasing the transverse mode number of the VCSEL would increase its group velocity difference as well as the relative time delay in MMF transmission due to the serious modal dispersion. In this work, an FM VCSEL with reduced transverse mode number can effectively minimize the modal dispersion of the data suffered during propagation in MMF, which helps to lengthen the allowable transmission distance with suppressed waveform distortion on the carried data stream for intra-data-center applications. In fabricating the FM VCSEL chip, a 350 μm thick GaAs wafer was cleaned by soaking in 95°C acetone for 5 min and followed by rinsing with isopropyl alcohol (IPA). After drying with nitrogen blow and baking at 125°C for 5 min, the ${\mathrm{Si}}_3{\mathrm{N}}_4$ layer was deposited and patterned as a hard mask for the reactive ion etching (RIE) shaping of the top DBR with heavy zinc diffusion. The zinc diffusion step was carried out before the process of p-type metal evaporation. In the zinc-diffused region of the top DBR, the induced disordering of ${\mathrm{Al}}_{0.9}{\mathrm{Ga}}_{0.1}\mathrm{As}$ and ${\mathrm{Al}}_{0.12}{\mathrm{Ga}}_{0.88}\mathrm{As}$ also improves the continuity of bandgap energy and refractive index to avoid free-carrier absorption. In this way, the resistance of the top DBR was effectively reduced to benefit from enhancing the modulation bandwidth. This wafer was provided from a commercial epi-foundry (LANDMARK Co.), and the doping type of the substrate is semi-insulating. Afterwards, a dilute hydrochloric acid solution was used to etch the oxide-covered GaAs layer for defining the p-type metal region. Later on, a Ti/Pt/Au metal was deposited as the p-type contact via E-gun evaporation, wherein the p-type mesa size and the wet oxidation depth help to determine the oxide-confined aperture of the FM VCSEL chip. For the FM VCSEL chip, photolithography was introduced to define the outer mesa diameter of 18 μm, and the RIE with ${\mathrm{SiCl}}_4/\mathrm{Ar}$ gaseous recipe was used to dry-etch the AlGaAs layer. The rate and depth of wet oxidation were controlled at 0.35 μm/min and 6.5 μm for achieving the oxide-confined aperture of 5 μm. A similar process was performed for defining the n-type contact with Au/Ge/Ni/Au deposition. Finally, a thick BCB layer was coated as the passivation layer to provide the low capacitance for the FM VCSEL chip.

B. On-Chip Test of 16-QAM OFDM Transmission

The on-wafer test of the 16-QAM OFDM directly modulated 850 nm FM VCSEL chip for 100 m OM4 MMF transmission is shown in Fig. 1. For on-chip driving of the VCSEL, a homemade probe station consisting of a coplanar ground–signal–ground (GSG) probe with a pitch (G–S spacing) of 100 μm was employed to provide an analog bandwidth beyond 40 GHz. The temperature of the VCSEL chip was precisely controlled at 22°C with a water-cooled heat sink set below the FM VCSEL chip, sealed upon the Si wafer with indium. A 65 GHz bias tee (Anritsu V250) was employed to combine the DC bias current with the transmitted 16 QAM OFDM. A homemade Matlab program was employed for encoding and decoding the 16-QAM OFDM data, in which a pseudo-random bit sequence (PRBS) data with a length of $2^{15}-1$ was mapped into 16-QAM symbols and serial-to-parallel distributed into all OFDM subcarriers in the frequency domain. Subsequently, the temporal OFDM waveform was generated through the inverse-fast-Fourier transform (IFFT) process with a fast Fourier transform (FFT) size of 512. The number of encoded subcarriers for the 23 and 25 GHz 16-QAM OFDM data at a sampling rate of 65 GS/s are 182 and 197, respectively, where each OFDM subcarrier is with a Baud rate of 127 MHz. Afterwards, electric 16-QAM OFDM data with a peak-to-peak amplitude of 1 V was exported from an arbitrary waveform generator (Keysight M8195A) with a sampling rate of 65 GS/s. The parameters of the 25-GHz 16-QAM OFDM data are shown in Table 1.

 

Fig. 1. (a) Experimental setup of proposed FM zinc-diffused VCSEL chip at 850 nm based on 16-QAM OFDM over 100 m MMF and (b) block diagrams of the encoding and decoding algorithms for 16-QAM OFDM transmission.

Download Full Size | PDF

Table 1. Information on the 16-QAM OFDM Para

View Table

To directly modulate the 16-QAM OFDM data stream, the DC bias of the FM VCSEL chip was optimized at 5 mA. For short-reach transmission in MMF, the optical data was first coupled into a lensed MMF that connects with an OM4 MMF patchcord with FC/PC connector. After 100 m long MMF (Corning ClearCurve OM4 fiber) transmission, a photodetector (PD, New Focus 1484-A-50) with a 3 dB bandwidth of 22 GHz was employed to convert the electrical 16-QAM OFDM data. To compensate the MMF transmission-induced power attenuation, the received data was resumed by a 35 GHz microwave amplifier (AMP, Picosecond 5882) with a gain of 16 dB and noise figure of 6 dB. Finally, the amplified electrical 16-QAM OFDM data was sent into a real-time digital serial analyzer (Tektronix DPO77002SX) with a sampling rate of 100 GS/s, and the waveform of the 16-QAM OFDM data was resampled for decoding analyses in an off-line, in-house-developed Matlab program.

In fact, the total bit length ($L_{\text{total}}$) of the PRBS patterns uploaded onto the OFDM data stream is given by $L_{\text{total}}=N_{\text{subcarrier}}\times N_{\text{symbol}}\times N_{\text{bit}}$, where $N_{\text{subcarrier}}$ denotes the subcarrier number, $N_{\text{symbol}}$ is the symbol number, and $N_{\text{bit}}$ is the bit per symbol. To synthesize the 16-QAM OFDM data stream covering a bandwidth of 25 GHz and sampling rate of 65 GS/s, the required $N_{\text{subcarrier}}$, Nsymbol, and $N_{\text{bit}}$ are 197, 2048, and 4, respectively. This leads to an $L_{\text{total}}$ of 1613824, which is close to a PRBS pattern length of $2^{21}-1$.

3. RESULTS AND DISCUSSION

A. Basic Characteristics of the FM VCSEL Chip

Figure 2(a) illustrates the schematic layer structure of the homemade FM VCSEL chip. The bottom n-type DBR layer of the FM VCSEL chip was made with 23 pairs of ${\mathrm{Al}}_{0.9}{\mathrm{Ga}}_{0.1}\mathrm{As}/{\mathrm{Al}}_{0.12}{\mathrm{Ga}}_{0.88}\mathrm{As}$ layers. The intrinsic active region consists of three 5 nm thick ${\mathrm{In}}_{0.08}{\mathrm{Ga}}_{0.92}\mathrm{As}$ quantum wells separated by two 8 nm thick ${\mathrm{Al}}_{0.35}{\mathrm{Ga}}_{0.65}\mathrm{As}$ barriers, which are sandwiched between two lightly doped graded layers (${\mathrm{Al}}_{0.35}{\mathrm{Ga}}_{0.65}\mathrm{As}$ and ${\mathrm{Al}}_{0.9}{\mathrm{Ga}}_{0.1}\mathrm{As}$) with thickness of 45 nm. For spatial gain confinement, an aperture layer consisting of carbon-doped ${\mathrm{Al}}_{0.98}{\mathrm{Ga}}_{0.02}\mathrm{As}$ with thickness of 60 nm and an oxide-confined aperture diameter of 5 μm was fabricated to strictly control the transverse mode number. Afterwards, the top of the FM VCSEL chip was completed by covering 21 pairs of lightly carbon-doped p-type ${\mathrm{Al}}_{0.9}{\mathrm{Ga}}_{0.1}\mathrm{As}/{\mathrm{Al}}_{0.12}{\mathrm{Ga}}_{0.88}\mathrm{As}$ DBR mirror layers, and then encapsulating them with a zinc-diffused p-type GaAs contact layer with a thickness of 50 nm. Note that the diameters of the top and bottom mesas of the FM VCSEL chip are 18 and 28 μm, respectively. The left inset of Fig. 2(a) shows the top-view and zoom-in microscopic images of the metallic pads of the FM VCSEL chips, which were designed with a coplanar waveguide contact configuration for connecting the coplanar GSG microprobes with a pitch spacing of 100 μm.

 

Fig. 2. (a) Schematic layer structure, (b) stimulated and (c) experimental optical spectrum, and (d) power-to-current and (e) voltage-to-current responses of the FM VCSEL chip.

Download Full Size | PDF

The basic characteristics of the FM VCSEL chip are shown in Figs. 2(b)–2(e). Figure 2(c) shows the optical spectrum of the FM VCSEL chip, which shows only three transverse modes around the central mode at a wavelength of 839.44 nm. Owing to the smaller oxide-confined aperture of 5 μm, the higher injection current density is ensured at lower threshold current for the FM VCSEL chip [26]. Figure 2(d) shows the power-to-current (P–I) response of the heavily zinc-diffused FM VCSEL chip, which reveals a threshold current of 0.22 mA. The 5 μm VCSEL chip endures some heat accumulation, which distinctly exhibits an output saturation trend under highly biased conditions, with 1 dB suppression beyond 25$I_{\mathrm{th}}$. On the other hand, the P–I response remains stabilized, even when delivering the highest optical power of 2.25 mW. As shown in Fig. 2(e), a differential resistance of 108.3 Ω with return loss of $-8.7\mathrm{dB}$ can be obtained at an optimized bias current of 5 mA with corresponding compliance voltage of about 2.7 V. When operating the FM VCSEL chip at compliance voltage beyond 2.5 V, its differential resistance response reveals a fluctuation trend due to the instability of the zinc-diffused p-type GaAs contact layer.

The relationship between transverse mode number and aperture size can be simply judged by calculating the character frequency of $V=\pi D{(n_{\text{core}}^2-n_{\text{clad}}^2)}^{0.5}/\lambda$, with D denoting the aperture size, $\lambda$ the wavelength, and $n_{\text{core}}$ and $n_{\text{clad}}$ the refractive indices of core and cladding, respectively. In principle, the definition of single transverse mode would include one TE mode and one TM mode. In a previous work with specified core and cladding refractive indices, the critical condition for a single-mode VCSEL with $V=2.405$ is obtained at $D\approx 2.6\mathrm{\mu m}$ [27,28]; that is, the VCSEL will emit more than one single transverse mode when the condition of $V>2.405$ is obtained with $D>3\mathrm{\mu m}$. Theoretically, the FM VCSEL with $D=5\mathrm{\mu m}$ will provide a transverse mode number of 4 (the total transverse mode number is $M\approx 4V^2/{\pi }^2$ including TE and TM modes), which is in a good agreement with our experimental observation. When the aperture size is decreased, the spacing of the transverse mode is increased. Hence, the overlap between the active layer and optical field of high-order modes is reduced accordingly. Therefore, the transverse mode number can be precisely controlled by the aperture size with high reproducibility.

To obtain fewer than five transverse modes, the simulation shows that the aperture size must be controlled within 5 μm, as shown in Fig. 2(b). The experimental result in Fig. 2(c) is in good agreement with this simulation result. In our work, the aperture size of 5 μm is a critical condition for fabrication of the FM VCSEL. If the aperture size is between 3 and 5 μm, both the number of transverse modes and the power of the VCSEL output will be further decreased, and will completely become single mode when aperture size decreases to 3 μm or less. In fact, the influence of these weak transverse mode-induced modal dispersions on the carried data is negligible, as their modal powers are too small to be taken into account. According to the IEEE P802.3 bs standard for VCSEL transmitters used in data centers, the root-mean-square (RMS) spectral linewidth ($\mathrm{\Delta }{\lambda }_{\mathrm{RMS}}$) of the VCSEL output is described as [29]

The normalized modulation output of the FM VCSEL chip is illustrated in Fig. 3(a). Obviously, the 3 dB modulation bandwidth is increased from 13.8 to 22 GHz by greatly increasing the bias current from 1.1 (5$I_{\mathrm{th}}$) to 4.4 mA (20$I_{\mathrm{th}}$). Note that the improvement on the available bandwidth of the FM VCSEL is mainly attributable to the zinc diffusion, which effectively reduces the top-DBR resistance to provide broadband modulation. In detail, the bias-dependent RIN responses of the FM VCSEL chip at different biases are measured using a lightwave signal analyzer (HP, 71300C) and compared in Fig. 3(b). Note that the RIN response is related to the relaxation oscillation frequency [33], and the RIN peak and power level are upshifted from 7.92 to 15.75 GHz and suppressed from $-154.4$ to $-163.1\mathrm{dBc}/\mathrm{Hz}$ after increasing the bias current from 1.1 to 4.4 mA, respectively.

 

Fig. 3. (a) Normalized modulation output and (b) RIN responses of the FM VCSEL chip.

Download Full Size | PDF

According to derivation in Ref. [34], the relaxation oscillation frequency ($f_r$) of the FM VCSEL can be expressed as

B. BtB 16-QAM OFDM Transmission of the FM VCSEL Chip

As to optimizing the bias condition, the subcarrier SNRs and the overall BER responses of the BtB transmitted 16-QAM OFDM data directly modulated on the FM VCSEL chip at detuned bias currents are shown in Figs. 4(a) and 4(b), respectively. For 16-QAM OFDM data within the 20 GHz provided by the FM VCSEL chip biased at 3 mA, an average SNR of 16.6 dB and BER of $1\times {10}^{-3}$ are achieved with a raw data rate of 80 Gb/s. By enlarging the bias current to 5 mA, the average SNR and BER are improved to 17.5 dB and $2.9\times {10}^{-4}$, respectively. This is attributed to the up-shifted RIN peak response as well as the increased 3 dB modulation bandwidth of the FM VCSEL at high bias. Conversely, continuous increment on the bias current to 5.5 mA deteriorates the average SNR and BER to 17.2 dB and $8.1\times {10}^{-4}$, respectively, resulting from the declined modulation power-to-frequency response and the unexpected roll-off phenomenon occurring in the low-frequency region [35].

 

Fig. 4. (a) Subcarrier SNR and BER responses, and (b) average BER of the 16-QAM OFDM data carried by the FM VCSEL at different bias currents.

Download Full Size | PDF

The peak-to-average power ratio (PAPR) is a dimensionless quantity defined as the ratio of instantaneous peaks to the average power of the OFDM waveform in the time domain, which can be expressed as [36]

Figure 6(a) shows the optimization of the FM VCSEL chip with the received BER, on modulating bandwidth, for the carried 16-QAM OFDM data with and without pre-leveling illustrated as a function of bandwidth while keeping the optimized DC bias at 5 mA. Without pre-leveling, the FM VCSEL chip enables maximal allowable OFDM subcarrier bandwidths up to 25 GHz for delivering total bit rate up to 100 Gb/s, with a spectral bandwidth usage efficiency of 4 bit/s/Hz. The constellation plots of the carried 16-QAM OFDM data at 25 GHz are shown in inset of Fig. 6(a).

 

Fig. 5. CCDFs of the PAPR for the QAM OFDM data carried by the FM VCSEL at different bias currents.

Download Full Size | PDF

 

Fig. 6. (a) BERs and constellation plots of 16-QAM OFDM data carried by the FM VCSEL without and with pre-leveling at different bandwidths. (b) Subcarrier SNR and BER of the 100 Gb/s 16-QAM OFDM data carried by the FM VCSEL chip.

Download Full Size | PDF

The corresponding BER, SNR, and EVM of $3\times {10}^{-3}$, 15.5 dB, and 16.8%, respectively, are observed. Pre-leveling the amplitude of OFDM subcarriers slightly suppresses the received BER to $2.9\times {10}^{-3}$ but maintains the maximal allowable 16-QAM OFDM data bandwidth at 25 GHz for the FM VCSEL chip. The SNR responses of 100 Gb/s data carried by the FM VCSEL without and with pre-leveling are shown in Fig. 6(b). Note that serious drops of the SNR response are observed at subcarrier frequencies of $<5$ and $>20\mathrm{GHz}$; these are induced by the insufficient response of the employed PD, and the dashed line in Fig. 6(b) illustrates the frequency response of the used PD.

C. 100 m OM4 MMF 16-QAM OFDM Transmission of the FM VCSEL Chip

For single-carrier intra-data-center applications, the transmission distance of the OM4 MMF is lengthened to 100 m for the purposes of performance evaluation. As a result, the SNR of the 16-QAM OFDM data carried by the FM VCSEL chip can cover total subcarrier bandwidth up to 23 GHz for a raw data rate of 92 Gb/s, as illustrated in Fig. 7(a). With the FEC overhead, the raw data rate of the 16-QAM OFDM data delivered by our VCSEL is 92 Gb/s. As the data carried by the FM VCSEL chip would still suffer from MMF-induced modal dispersion, the average SNR of the 16-QAM OFDM data somewhat decreases from 15.9 to 14.8 dB after transmission in 100 m MMF. Figure 7(b) shows the constellation plots of the 16-QAM OFDM data carried by the FM VCSEL chip before and after 100 m OM4 MMF transmissions. The plots become blurred, revealing the deteriorating EVM from 16% to 18% and the enlarged BER from $1.9\times {10}^{-3}$ to $5\times {10}^{-3}$. To improve the receiving performance, the QAM-OFDM data is pre-compensated by pre-leveling its subcarrier amplitude to slightly sacrifice the low-frequency SNR, essentially enhancing the SNRs at high frequencies.

 

Fig. 7. (a) Subcarrier SNR and BER responses and (b) constellation plots of the 16-QAM OFDM data before and after 100 m MMF transmissions.

Download Full Size | PDF

Later, the OFDM subcarrier pre-leveling technique we use in our program is in fact the pre-equalization, which can resist the declining frequency response of the VCSEL and extend the allowable OFDM data bandwidth for maximizing the transmission capacity [33]. This pre-leveling technique is illustrated in Fig. 8. After implementing the OFDM subcarrier pre-leveling, the power of the electrical OFDM subcarriers is preset through multiplication with an exponential, increasing function in frequency domain before IFFT, which can use the low-frequency OFDM subcarrier powers to compensate for the high-frequency power degradation. This effectively improves the SNR of the received OFDM data.

 

Fig. 8. OFDM data after transmission (a) without and (b) with pre-leveling technique.

Download Full Size | PDF

Figure 9(a) shows the effect of the subcarrier pre-leveling on the BER of the 92-Gb/s 16-QAM OFDM data carried by the FM VCSEL chip. With an amplitude pre-leveling slope of 0.3 dB/GHz, the BER of the carried QAM-OFDM data effectively improves from $5\times {10}^{-3}$ to $3.6\times {10}^{-3}$ ($<\mathrm{FEC}$ criterion of $3.8\times {10}^{-3}$). The SNR response of the QAM-OFDM data shown in Fig. 9(b) can be improved by $\sim 1.5\mathrm{dB}$ to exceed 15.3 dB for the FEC requirement. The constellation plots shown in Fig. 9(c) also confirm that the EVM of the decoded QAM-OFDM data can be suppressed from 18% to 17.2% after pre-leveling. Over pre-leveling, the subcarrier amplitude with slope beyond 0.4 dB/GHz excessively sacrifices the low-frequency SNR without further compensating the high-frequency SNR, eventually reducing the average SNR. After BtB and 100 m MMF transmissions, the BER curves of the 92 Gb/s 16-QAM OFDM data carried by the FM VCSEL chip without and with subcarrier pre-leveling are compared in Fig. 9(d). To meet the FEC criterion, the BtB transmitted QAM-OFDM data without pre-leveling requires a receiving power sensitivity of at least $-4\mathrm{dBm}$, which inevitably increases up to $\gg 1\mathrm{dBm}$ after propagation over 100 m OM4 MMF. With subcarrier pre-leveling, the receiving power penalty of the QAM-OFDM data between BtB and 100 m MMF transmissions can be efficiently suppressed to 4 dB.

 

Fig. 9. (a) Average BER, (b) subcarrier SNR and BER responses, and (c) constellation performance of the 16-QAM OFDM data transmitted over 100 m MMF at different pre-leveling slopes. (d) BER versus receiving power of the 16-QAM OFDM data carried by the FM VCSEL with and without pre-leveling under BtB and 100 m MMF transmissions; Pre: pre-leveling.

Download Full Size | PDF

For implementing the 100 Gb/s transmission, the previous work used an 850 nm VCSEL with an RIN of $-145\mathrm{dBc}/\mathrm{Hz}$ to achieve PAM-4 data transmission over 100 m long MMF [21]. In fact, the VCSEL only supports up to a 25 Gb/s data rate in OOK format, and the bit-rate improvement with PAM-4 format relies strictly on pre-emphasis technology with high-speed amplifier, Oeder–Meyer timing recovery, pre-emphasis filter, and adaptive equalizer. In contrast, our proposed FM VCSEL chip exhibits significantly low RIN of −$163.1\mathrm{dBc}/\mathrm{Hz}$ to enable 16-QAM OFDM data with high criterion on SNR, which benefits from reduced power consumption during amplification within finite modulated bandwidth. To compare, this work proposes an FM VCSEL chip with a smaller oxide-confined aperture to show a competitive transmission capacity, which guarantees its potential for applications with lengthened distance.

4. CONCLUSION

An FM VCSEL chip with heavily zinc-diffused contact layer and oxide-confined reduced cross-section area is demonstrated, carrying pre-leveled 16-QAM OFDM data at 100 Gb/s for transmission over 100 m in OM4 MMF. The FM VCSEL reveals a differential resistance of 108.3 Ω with return loss of $-8.7\mathrm{dB}$ when providing the highest optical power of 2.25 mW, as obtained with delivering a stabilized P–I response at an optimized bias current of $>4.4\mathrm{mA}$ with compliance voltage of 2.7 V. At 4.4 mA, the FM VCSEL chip supports a 3 dB modulation bandwidth of 22 GHz with RIN peak as low as $-163.1\mathrm{dBc}/\mathrm{Hz}$. For data modulation, the FM VCSEL chip enables maximal allowable OFDM data bandwidths up to 25 GHz, carrying the bit rate up to 100 Gb/s with corresponding BER, SNR, and EVM of $3\times {10}^{-3}$, 15.5 dB, and 16.8%, respectively. For practical intra-data-center applications, the transmission distance in the OM4 MMF can be lengthened to 100 m to pass the FEC criterion, and the FM VCSEL chip is capable of carrying 16-QAM OFDM data with bandwidth up to 23 GHz, for a raw data rate of 92 Gb/s. With subcarrier amplitude pre-leveling, the receiving power penalty of the data received between BtB and 100 m MMF transmissions is efficiently suppressed to 4 dB. In this work, it is determined that an FM VCSEL with reduced transverse mode number can effectively minimize the modal dispersion of data during MMF transmission, which helps to lengthen the allowable transmission distance with suppressed waveform distortion on the carried data stream. By employing the OFDM subcarrier pre-leveling to further improve the 16-QAM OFDM data carried by the FM VCSEL, an FEC-certified transmission of 92 Gb/s over 100 m in OM4 MMF is realized to facilitate data center or OI applications.